• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

360 Gpu Exposed, 4XMSAA etc , and PS3GPU was a late change.

Pug

Member
http://techreport.com/etc/2005q2/xbox360-gpu/index.x?pg=1

96 billion shader ops confirmed. On chip, the shaders are organized in three SIMD engines with 16 processors per unit, for a total of 48 shaders. Each of these shaders is comprised of four ALUs that can execute a single operation per cycle, so that each shader unit can execute four floating-point ops per cycle. 4XMSAA is virtually free less than 5% hit. I have no doubts unified shaders are the future of GPUs.

Sorry here the link to the info regarding the G70 http://www.firingsquad.com/features/xbox_360_interview/default.asp
 
I posted this at another thread but it didn't seem to be getting much mention. If true and accurate, I'd really like to get some opinions as I'm pretty clueless about these things.

block.gif


On chip, the shaders are organized in three SIMD engines with 16 processors per unit, for a total of 48 shaders. Each of these shaders is comprised of four ALUs that can execute a single operation per cycle, so that each shader unit can execute four floating-point ops per cycle.

500*48*4=96 Shader Ops....Is this correct?
 
it's mentioned here
http://www.firingsquad.com/features/xbox_360_interview/default.asp

It also says (also mentioned in the other thread)

FiringSquad: Do you know if it supports dual HD displays?

ATI: No it doesn’t. I know the NVIDIA chip does, but that’s only because PC products do. It doesn’t seem to have a real use inside the living room, but maybe you differ with me on that.

FiringSquad: Well, on the Sony console, I think they’re looking at applications that go beyond just a console in the living room don’t you think?

ATI: Yeah I really think it’s just an accident because, well you know, last summer they had to change their plans. They found out that Cell didn’t work as well as they wanted to for graphics. Remember originally you had two or three Cell processors doing everything and then in August last year they had to take an NVIDIA PC chip. And as you know, all PC chips do this, and so it [dual HD display outputs] just came for free.

something to think about.
 
By all accounts
PS3 GPU == next-gen pc nvidia gpu chip + a lot of power
X360 GPU == custom gpu with a lot of cool new features + a lot of power as well
 
Mr Bob, it just points out what many of us suspected. The switch to PC G70 was taken late, and the PC G70 will be very simialr to the GPU in PS3.
 
ATI: Yeah I really think it’s just an accident because, well you know, last summer they had to change their plans. They found out that Cell didn’t work as well as they wanted to for graphics. Remember originally you had two or three Cell processors doing everything and then in August last year they had to take an NVIDIA PC chip. And as you know, all PC chips do this, and so it [dual HD display outputs] just came for free.

Oh, we have a ATI guy who is a Sony/nVidia insider.

I thought nVidia has been working for a good 2 years with Sony.


and the PC G70 will be very simialr to the GPU in PS3.

Isn't the G70 about to be introduced?

Cool. Now we will see what Sonys GPU is all about since this is confirmed by a ATI insider within Sony/nVidia!
 
I thought it was pretty common knowledge that nvida was a late addition to ps3. When it was announced the BD3 forum went crazy because Sony was going to get a real graphic chip maker to the design the gpu to go along with the cell.
 
Pug said:
Mr Bob, it just points out what many of us suspected. The switch to PC G70 was taken late, and the PC G70 will be very simialr to the GPU in PS3.
It could be true, but this is also based on anecdotal evidence. For one, no one knows what the internal makeup of the RSX is. As Phil Harrison mentioned, NVidia ain't making the chip. RSX is a Sony chip with NVidia IPs in it. Which begs the question, what else did they put on-board. But besides that, what about the G70's architecture suggests that it's just a PC part, lack of eDRAM? And is it really the G70, or the G80? Lastly, isn't the Gxx line now a new line of GPUs from NVidia, different from the previous lines? I thought the chip was a departure from the stuff they had in the 6800's. Fill me in, b/c I haven't seen anything detailed on the RSX yet to suggest that it's a straight PC part. Lack of eDRAM means nothing since that was speculated long before the final product was shown. PEACE.
 
AB101. if they had been working for 2 years with a hundred engineers then we may have seen a pixel shader GPU and cell blast out the vertex calculations. It was a late change Pani probably know more though. Pimp, the G70 Comparisons are just from the rumours from the linked sites, they seem pretty confident, but Sony will certaintly have made sure its work well in tandom with Cell.
 
thorns said:
By all accounts
PS3 GPU == next-gen pc nvidia gpu chip + a lot of power
X360 GPU == custom gpu with a lot of cool new features + a lot of power as well

So basically this means that both chips are pretty close in power, and the Cell has a bit more power than X360's CPU. Both systems seems pretty evenly matched. Nice.

It's all down to the games, as usual, but hopefully there'll be less fanboy "your system can't handle this shit!" next-gen :)
 
We were worried, you know obviously when you integrate the unified shaders and the EDRAM, we were a bit worried that there could be efficiency problems in the unified shaders, that there could just be general problems with EDRAM.

That's what I'd be worried about too. I dont know the details about shaders but normally when you merge 2 seperate things into one there's going to be some kind of penalty, like more cache misses or something. Pana?
 
So this means the X360 should be able to accomplish something close to the Killzone rendered demo? Sweet! :D X360 ONE CONSOLE FUTURE.
 
Mrbob said:
So this means the X360 should be able to accomplish something close to the Killzone rendered demo? Sweet! :D X360 ONE CONSOLE FUTURE.

It is amazing how many EE's we have in here that are so much smarter than Sony, Microsoft, ATI, & NVidia. ;)

Don't any of you think that if Microsoft had any sort of semblance of power advantage that they'd be trumpeting it this week that they've taken such a beating? Or that Sony would be leary to say that their system was 2x as powerful as the Xbox360?

Microsoft has been called out on numerous occasions and they continually say, "it's not about the hardware, it's about the software and services".

And don't even try to tell me that both Sony & Microsoft aren't intimately aware of what each other has hardware wise w/ respect to the CPU & GPU. But, if you want, keep the hope alive. Maybe it'll come true, maybe it won't. Maybe one of Microsoft or Sony don't really know what type of system they have and have been "found out" by the experts of the message boards.
 
AB 101 said:
Oh, we have a ATI guy who is a Sony/nVidia insider.

I thought nVidia has been working for a good 2 years with Sony.

you dont think ATI would have been offered a bid by sony along with NV?
 
Don't any of you think that if Microsoft had any sort of semblance of power advantage that they'd be trumpeting it this week that they've taken such a beating? Or that Sony would be leary to say that their system was 2x as powerful as the Xbox360?

Microsoft has been called out on numerous occasions and they continually say, "it's not about the hardware, it's about the software and services".
Did you not watch the MS conference? I could count at least 3 times where they called X360 the most powerful console :)
 
sonycowboy said:
It is amazing how many EE's we have in here that are so much smarter than Sony, Microsoft, ATI, & NVidia. ;)

Don't any of you think that if Microsoft had any sort of semblance of power advantage that they'd be trumpeting it this week that they've taken such a beating? Or that Sony would be leary to say that their system was 2x as powerful as the Xbox360?

Microsoft has been called out on numerous occasions and they continually say, "it's not about the hardware, it's about the software and services".

And don't even try to tell me that both Sony & Microsoft aren't intimately aware of what each other has hardware wise w/ respect to the CPU & GPU. But, if you want, keep the hope alive. Maybe it'll come true, maybe it won't. Maybe one of Microsoft or Sony don't really know what type of system they have and have been "found out" by the experts of the message boards.
quoted for truth

also, why are we dive-waving? It's like "my 80 foot long dick is bigger than your 75 foot long dick"

SONY AND MICROSOFT BOTH HAVE REALLY BIG DICKS THIS GEN DON'T WORRY

everybody = teh happy
 
sonycowboy said:
Do you think that if Microsoft had any sort of semblance of power advantage that they'd be trumpeting it this week that they've taken such a beating? Or that Sony would be leary to say that their system was 2x as powerful as the Xbox360?

Sony has no problem lying their ass off. What happened to that 10 Teraflop total performance number they chucked out? :lol
 
Pug said:
AB101. if they had been working for 2 years with a hundred engineers then we may have seen a pixel shader GPU and cell blast out the vertex calculations. It was a late change Pani probably know more though. Pimp, the G70 Comparisons are just from the rumours from the linked sites, they seem pretty confident, but Sony will certaintly have made sure its work well in tandom with Cell.
AFAIK, NVidia and Sony announced that it was a team of 50 engineers from NVidia, not hundreds...just 50.

Source: X-bit labs interview w/ David Roman
Anna (X-bit labs): As far as I know NVIDIA may claim that there was not that much investment into the RND: only about 50 engineers. Is this the result of the fact that Sony’s own engineers contributed to the development of the GPU in a significant way?

David Roman: We do not disclose anything on the actual resources. Obviously there is a major economy of scale. This chip is a custom version of our next generation GPU. So we’ve been working on the next generation GPU for close to two years now, namely about 18 months. I don’t know the cost of this one but I know the cost of the last generation: it was 350 million dollars. These are expensive chips to develop. So, the fact that we didn’t have to do that development just for the Sony application obviously is a major economy of scale, because we are doing the development for the new chip anyway. The amount of work involved into customization, I don’t know. I know that we designed a new generic team, we had been working with Sony before on the actual development platform, we had actually been working on the details of the chip. We now have assigned an engineering team to work as a Sony engineering team. And the numbers? I don’t know what the numbers are but I am sure they are growing, but there is a lot of work that’s going on. As I have said we do not disclose the details, but there is certainly some economy of scale due to building it on the technology that we have been working on for a long time. So, it is the next generation of GPU.
:D I'm grinning b/c I rarely source any of my claims, but I figured if I want to avoid looking like I'm just running damage control, I better bring some evidence this time. :lol I'd even made a post about it over at B3D at the time, b/c I found it puzzling that so few people would be working on a rushed GPU project. Doesn't seem like it was so rushed if this is what the final product was. And it corroborates Phil's claim that this is really a Sony part.

I'm still hoping for some Chaperone stuff, or something like it, b/c that's essentially what ATI put in Xenos, some self-shadowing and AA hw that frees up bandwidth and the GPU. And with "amatuer" techies at B3D instantly questioning the bandwidth comparison between Xenos and RSX, I would hope NVidia/Sony had an answer to this. That eDRAM on Xenos is easily the coolest thing about the architecture IMO. I don't think unified ALUs will be the future. I don't think that's a given at all. But that eDRAM clearly gives the chip a nice bump in efficiency. PEACE.
 
Sony cowboy, IF these figures are correct and they came from an interview with ATI the 360 GPU can do 192op/clock and with free 4xmsaa to boot. This makes it quite a bit more powerful than the RSX Gpu. But as always I'm sure there other stuff we don't know yet. However its pretty safe to say the 360 will be at LEAST as powerful as the RSX and probably more so. And pimp I said it was speculation but it does say in the article a custom based G70. By the have you read the ATI interview?

"We have 48 shaders. And each shader, every cycle can do 4 floating-point operations, so that gives you 196.

The 192 is actually in our intelligent memory, every cycle we have 192 processors in our embedded intelligent memory that do things like z, alpha, stencil. So there are two different numbers and they’re kind of close to each other, which leads to some confusion. "
 
FiringSquad: How does Xbox 360 GPU compare in size to the RSX?

ATI: In terms of size, we’re a bit smaller. Of course, I’m not sure if that’s a good way to compare things, and to be honest I can’t talk about the number of transistors for this design. Microsoft owns the IP and that has a lot to do with their cost model and all that sort of stuff. But we’re a very efficient engine and we feel very good about our design. You know, the bang for the buck is awesome. The power of the platform [pauses] we’re going to be the most powerful platform out there, we’ve got a lot of innovation in there, we’re not just a PC chip.

I think the Sony chip is going to be more expensive and awkward. We make efficient use of our shaders, we have 64 threads that we can have on the processor at once. We have a thread buffer inside to keep the [inaudible]. The threads consist of 64 vertices or 64 pixels. We have a lot of work that we can do, a lot of sophistication that the developer never has to see.

I definitely believe this statement. In the PC arena ATI cards seem to be much more efficent than Nvidia offerings.
 
Pug said:
Sony cowboy, IF these figures are correct and they came from an interview with ATI the 360 GPU can do 192op/clock and with free 4xmsaa to boot. This makes it quite a bit more powerful than the RSX Gpu. But as always I'm sure there other stuff we don't know yet. However its pretty safe to say the 360 will be at LEAST as powerful as the RSX and probably more so. And pimp I said it was speculation but it does say in the article a custom based G70. By the have you read the ATI interview?

"We have 48 shaders. And each shader, every cycle can do 4 floating-point operations, so that gives you 196.

The 192 is actually in our intelligent memory, every cycle we have 192 processors in our embedded intelligent memory that do things like z, alpha, stencil. So there are two different numbers and they’re kind of close to each other, which leads to some confusion. "

Whoa.
 
Even if the PS3GPU is a little more powerful, what real world performance will it make? It seems like we are comparing an X800XT to a 6800ULtra, but the ATI may actually have better shader routines. All the stuff built into the embedded 10MB ram sounds cool too. I'll take MS setup AND a built in hdd with my console than the Nvidia setup and no HDD (I'm betting Sony will sell it seperately).
 
Pug said:
Sony cowboy, IF these figures are correct and they came from an interview with ATI the 360 GPU can do 192op/clock and with free 4xmsaa to boot. This makes it quite a bit more powerful than the RSX Gpu. But as always I'm sure there other stuff we don't know yet. However its pretty safe to say the 360 will be at LEAST as powerful as the RSX and probably more so. And pimp I said it was speculation but it does say in the article a custom based G70. By the have you read the ATI interview?

"We have 48 shaders. And each shader, every cycle can do 4 floating-point operations, so that gives you 196.

Dude, come on. If you're going to inflate numbers, then it's R500s 196FLOP/sec in it's shaders against RSX's 384FLOP/sec in it's.

And using "Shader Ops" is such bullshit, it's one of the most asymmetic metrics possible between different IHVs and architectures.
 
Wait....in this interview it says the ATI chip has an on die TV encoder for the 360. Does that mean....bigger hdd, TIVO like stuff I can do? I'm a little fuzzy on this part.
 
Vince I have no idea how you calculated the Nividia shader ops number. Still seems to me The 360 GPU certaintly matches up to the RSX in many areas. And off course its difficult to compare ATI unified shaders but thats not going to stop people.
 
Vince said:
Dude, come on. If you're going to inflate numbers, then it's R500s 196FLOP/sec in it's shaders against RSX's 384FLOP/sec in it's.
Shhh.... I wanted to see if anyone noticed that the numbers were off. :lol

To be honest, Anandtech is way off in that the Xenos/R500 has way more info out there about it than the RSX/G70. The number of shader ops and FLOPS were figured out weeks/months ago. As Vince said, the numbers are off. Not like it matters, all will be known in the fullness of time.

The RSX should be more powerful...it's coming out later, duh. That doesn't mean that eDRAM on Xenos isn't cool as shit though. I'd like to know how HDR affects performance on Xenos compared to RSX. From that Sony presser, it would seem that RSX was built from the ground up to do HDR. It was a major focus of the presentation afterall. But little is known about how RSX will handle the bandwidth requirements, and even less is known of how Xenos will handle it.

It's still early in this next-gen pissing match. I mean, only a few days now since the official unveiling, right? We can't expect them to blow their wad all at once. I think the next GDC should be REALLY interesting. :) PEACE.
 
Mrbob said:
Wait....in this interview it says the ATI chip has an on die TV encoder for the 360. Does that mean....bigger hdd, TIVO like stuff I can do? I'm a little fuzzy on this part.

I remember reading somewhere about the idea that you'd be able to get game invites and emails while watching TV over Xbox Live on the 360. I figure they probably need a TV encoder to do that. Not sure why they would NOT include some TIVO functionality if that is indeed the case.
 
GhaleonEB said:
Sony has no problem lying their ass off. What happened to that 10 Teraflop total performance number they chucked out? :lol

They didn't say that. At the PC, they were talking about a supercomputer that's at Sony Imageworld in Los Angeles and they showed a picture of it with '10 TFLOPS For Movie Rendering' on it.

After that Kutaragi said they were gonna use 'such Supercomputer-like performance in the real-time entertainment arena' and then a picture is shown with '2 TFLOPS for PlayStation 3' on it.
 
I already wrote something about shading operations here and on B3D, but I'm going to repeat it another time ;)
Shader ops are a MEANINGLESS unit measure, cause every hw vendor has different defintions of shader ops, even between different GPU generations from the same vendor!
If we want to try to compare different GPU shading power we should count floating point operations instead of shader operations.
Regarding R500: each ALU can do a vec4 operation and a scalar operation per clock cycle.
ATI says those are 2 shader operations (even if those 2 ops are COMPLETELY different things from a computational standpoint!), so 48 ALUs * 2 shader ops = 96 shader ops per cycle.
But we're smarter than them so we're going to count floating point operations per clock cycle.
R500's ALU does 10 floating point operations per cycle (8 ops from a vec multiply-add and 2 ops from a scalar multiply-add), so it's rated at 10*48*500 Mhz = 240 Gigaflop/s (this is a lot!)
What about RSX? Well..we don't know much about it. Nvidia released a couple of numbers:
1) 136 shader ops per cycle
2) 51 Giga dot products per second.

The first number is useless cause we don't know RSX ALUs, and we don't know how nvidia count shader ops. (remember: each vendor has its shader ops definitions)
The second number is someway interesting, it tells us RSX does 51*10^9/550*10^6 = 92 dot products per clock cycle.
R500 ALUs should be able to do one dot product per clock cycle, so RSX is almost 2x faster than R500 in this (frequently used) mathematical operation.
A dot4 takes 7 floating point ops, so we can tell RSX is rated at least at 350 Gigaflop/s,
but we can expect each RSX ALU to be able to do a dot product or a fmadd instruction (this is a very common thing in modern GPUs) so RSX rating goes up to 92*8*550 Mgz = 409 Gigaflop/s..wooow! :)

Disclaimer: I'm not saying those numbers are correct cause I extrapolated a lot of thing and made assumptions here and there, but please...just stop to use shading ops as an indicator of how much powerful is a GPU :)
 
Enough of the fanboy musings.

My concern is what effect will having one of the 360 cores handling audio and what effect will having the gpu act as the memory controller have on the performance of the system. Also another concern is the efficiency level of the buffer that feeds the shader units.
 
Nostromo said:
Disclaimer: I'm not saying those numbers are correct cause I extrapolated a lot of thing and made assumptions here and there, but please...just stop to use shading ops as an indicator of how much powerful is a GPU :)
Sure as long as you stop extrapolating and making assumptions. :)
 
"6X BRD-ROM drive"

Can anyone clarify this - in some Japanese news sources, they are saying this refers to the fact that the standard Blu Ray is around 6 times the capacity, NOT THE SPEED.

This wouldn't be the first time that sony has goofed up on drive information (recalls the PSX "12x DVD burner")
 
Where how and why are they suddenly claiming 96bn shader ops? People are saying the numbers are "off" - can anyone explain why?
 
DCharlie said:
"6X BRD-ROM drive"

Can anyone clarify this - in some Japanese news sources, they are saying this refers to the fact that the standard Blu Ray is around 6 times the capacity, NOT THE SPEED.

This wouldn't be the first time that sony has goofed up on drive information (recalls the PSX "12x DVD burner")
Thanks for pointing that out, I still thought that number was the speed as it was mentioned in the 'official Sony PC thread'. Bummer, but too good to be true I guess. :(

The speed isn't mentioned anywhere, it seems, not in the press release, not in the press conference. Will it 'just' be a 1x-drive or could it still be undecided?
 
Vince said:
Marco, has there been any word on how effecient ATIs scheduling/arbitration is?
No, but I bet is very efficient..;)
Maybe RSX wil be more powerful (as it probably performs more work per clock cycle) than R500, but R500 is probably way more elegant and it might be more efficient than NVIDIA solutions for any given transistor.
 
Oh wait, i think i see where the number's coming from now. 1 vector shader op = 4 floating point ops, right? For a 4-component vector? So they're just terming things in terms of flops now, not shader ops? The mention of 4 alus is what confused me...but i think that means 1 4-component vector alu.
 
gofreak said:
Where how and why are they suddenly claiming 96bn shader ops? People are saying the numbers are "off" - can anyone explain why?
Check marco's post. 96 shader ops/clock is derived from 48 pipes x 2 shader ops per pipe. So 96 shader ops x 500MHz = 48 billion shader ops per second, not 96.

I think we're confusing the 192 floating point ops from the eDRAM with the shader ops from the GPU. And I assume all those floating point ops allow for the free AA for all those pipes. Just an assumption, I'm not as well-versed in this as others. It helps to understand the programming side a lot IMO. I am almost worthless in that regard. ;) But I can find my way around a block diagram just fine. :)

BTW, there was never a drive speed given. 6x is the capacity as noted. I *hope* we see a 2x BRD drive. Because a 1x drive is slower than a 12x DVD, and that would mean...loading...loading...loading...loading. I'm hoping to finally play a Madden game without 30 seconds of loading. :( PEACE.
 
Pimpwerx said:
I think we're confusing the 192 floating point ops from the eDRAM with the shader ops from the GPU. And I assume all those floating point ops allow for the free AA for all those pipes. Just an assumption, I'm not as well-versed in this as others. It helps to understand the programming side a lot IMO. I am almost worthless in that regard. ;) But I can find my way around a block diagram just fine. :)

I think i figured it out above. It can do 48 vector ops and 48 scalar ops simultaneously (96 shader ops per cycle). A vector op is 4 floating point ops. So 48*4 = 192flops per cycle. This is a different metric than shader ops, which is still 48bn per second (96 * 500mhz).
 
Top Bottom